Unleashing the Power of AI to Transform PCB Design
May 31, 2024
One of the simplest ways to start artificial intelligence (AI) assisted PCB designing is to simply register on the CELUS Design Platform at: app.celus.io.
The first step is that you are asked to complete a Project Summary which includes a description of your project, the selection of the functionalities it should contain, the intended application, which CAD Tool should the project be handed over to as well as the possibility to determine preferred and/or excluded parts and manufacturers. The Project Settings stage has two particularly important functions. Firstly, it causes the user to pause and take a step back to think about what they want to do before launching themselves blindly into the software.
Secondly, it informs the platform about the essential parameters of the project so that it can tailor its advice and replies to better suit the project goals. The CELUS Design Platform was developed with artificial intelligence in mind from the very beginning, acting in many ways like a senior design engineer offering advice and knowledge to the next generation of design engineers, who may be bursting with ideas but simply lack the experience gained over many decades in the business.
It was this “companion” approach to project design and planning that attracted RECOM to be a partner with CELUS from the beginning. We could see the advantages of artificial intelligence when used as a time-saving tool –eliminating the drudgery of collating information, generating BoMs (bills of materials), creating netlists, and trawling through endless datasheets trying to find essential information such as efficiency figures, dimensions, or tolerances – work that could be safely assigned to a tireless AI assistant without giving the design engineer a feeling that they were no longer in control. However, in the intervening years, AI has moved onwards, and it now offers more than just assistance – namely collaboration.
For example, with the CELUS platform, once past the Project Settings and into the design stage, the software uses a familiar drag-and-drop style to create the system architecture block diagram. However, the lines linking the functional blocks could be power or data or both. It is not necessary to specify the connection type because the system understands how the functional blocks need to be interconnected. However, if the circuit designer has a particular preference, say, for an I2C data connection because they already have an existing interface firmware solution for that data type, then they can simply tell the system that that is what they want. The system will then choose the requisite interface when the schematic is generated.
This integration of artificial intelligence in design platforms heralds a paradigm shift in PCB design because, unlike conventional PCB software, which merely flags design rule violations, AI-powered platforms offer a transformative approach. AI enables the system to leverage vast databases of information with ease, coupled with the intelligence to suggest informed solutions, effectively translating project goals into functional electronic designs. Therefore, RECOM is in the process of integrating our product portfolio, which includes around 30.000 parts into the CELUS knowledge database. By tapping into this wealth of data, the AI can make nuanced component selections tailored to the specific requirements of each project, thereby enhancing efficiency and optimizing performance.
Despite the undeniable potential of AI in PCB design, it's natural for engineers to harbor concerns about its implications. Questions about job security and accountability often arise: Will AI take my job away? Will I be blamed if it makes a mistake? However, rather than being a threat, an AI assistant can serve as a dependable partner, capable of explaining its decisions and providing valuable insights. Its ability to justify choices fosters a collaborative environment where less-experienced engineers can learn and grow without feeling intimidated. Moreover, AI's capacity for continual learning means that it evolves alongside its users, constantly improving and adapting to new challenges.
These are the common design options that the layouter is familiar with and allows the user to take advantage of the head start given by the platform-generated prototype to make a fast time-to-market solution using their own custom design rules and preferences rather than having to use default settings. This handover process also optimizes the abilities of the different software platforms – AI is great to quickly turn an idea into a design, but the many specialized and advanced EDA platforms are ideal for generating the Gerber files containing the required CAM physical data such as the copper layers, solder masks, NC drill data, etc. Each to its own.
The boundary between the AI-assisted design and the layout software is not fixed. As the power of the machine learning algorithm increases, then more preparatory work can be done before handover. For example, when laying out a power electronics PCB, online calculators often need to be used to check the current capacity limits of tracks and vias. Existing EDA programs often have modules that can generate useful current density maps but can only make automatic changes to the layout if the voltage levels and component power demands are known. Thus, this part of the design process remains manual and relies heavily on the skill and experience of the designer to choose appropriate track widths and via aspect ratios. However, if this power consumption information could be made available to the artificial intelligence design assistant, this data could be synchronized with the layout software, so that machine-to-machine communication could be used to optimize the layout design automatically. Although such capabilities are not yet realized, ongoing advancements suggest they could become standard features soon.
Secondly, it informs the platform about the essential parameters of the project so that it can tailor its advice and replies to better suit the project goals. The CELUS Design Platform was developed with artificial intelligence in mind from the very beginning, acting in many ways like a senior design engineer offering advice and knowledge to the next generation of design engineers, who may be bursting with ideas but simply lack the experience gained over many decades in the business.
It was this “companion” approach to project design and planning that attracted RECOM to be a partner with CELUS from the beginning. We could see the advantages of artificial intelligence when used as a time-saving tool –eliminating the drudgery of collating information, generating BoMs (bills of materials), creating netlists, and trawling through endless datasheets trying to find essential information such as efficiency figures, dimensions, or tolerances – work that could be safely assigned to a tireless AI assistant without giving the design engineer a feeling that they were no longer in control. However, in the intervening years, AI has moved onwards, and it now offers more than just assistance – namely collaboration.
For example, with the CELUS platform, once past the Project Settings and into the design stage, the software uses a familiar drag-and-drop style to create the system architecture block diagram. However, the lines linking the functional blocks could be power or data or both. It is not necessary to specify the connection type because the system understands how the functional blocks need to be interconnected. However, if the circuit designer has a particular preference, say, for an I2C data connection because they already have an existing interface firmware solution for that data type, then they can simply tell the system that that is what they want. The system will then choose the requisite interface when the schematic is generated.
This integration of artificial intelligence in design platforms heralds a paradigm shift in PCB design because, unlike conventional PCB software, which merely flags design rule violations, AI-powered platforms offer a transformative approach. AI enables the system to leverage vast databases of information with ease, coupled with the intelligence to suggest informed solutions, effectively translating project goals into functional electronic designs. Therefore, RECOM is in the process of integrating our product portfolio, which includes around 30.000 parts into the CELUS knowledge database. By tapping into this wealth of data, the AI can make nuanced component selections tailored to the specific requirements of each project, thereby enhancing efficiency and optimizing performance.
Despite the undeniable potential of AI in PCB design, it's natural for engineers to harbor concerns about its implications. Questions about job security and accountability often arise: Will AI take my job away? Will I be blamed if it makes a mistake? However, rather than being a threat, an AI assistant can serve as a dependable partner, capable of explaining its decisions and providing valuable insights. Its ability to justify choices fosters a collaborative environment where less-experienced engineers can learn and grow without feeling intimidated. Moreover, AI's capacity for continual learning means that it evolves alongside its users, constantly improving and adapting to new challenges.
So, what can Artificial Intelligence in PCB design do, almost do, and not yet do?
Software platforms such as CELUS take the block diagram and find suitable solutions for circuit designers to evaluate and generate the schematic, BoM (bill of materials), floorplan proposal and footprints in a choice of different Electronic Design Automation (EDA) formats that are compatible with popular PCB Layout software such as Altium Designer, Autodesk Eagle, and KiCad. Once in the chosen native EDA format, the user can further modify the given solution to optimize the design, such as changing the component placement, adding polygons or copper pour to fill in the planes, setting component groups, changing the stack up, etc.These are the common design options that the layouter is familiar with and allows the user to take advantage of the head start given by the platform-generated prototype to make a fast time-to-market solution using their own custom design rules and preferences rather than having to use default settings. This handover process also optimizes the abilities of the different software platforms – AI is great to quickly turn an idea into a design, but the many specialized and advanced EDA platforms are ideal for generating the Gerber files containing the required CAM physical data such as the copper layers, solder masks, NC drill data, etc. Each to its own.
The boundary between the AI-assisted design and the layout software is not fixed. As the power of the machine learning algorithm increases, then more preparatory work can be done before handover. For example, when laying out a power electronics PCB, online calculators often need to be used to check the current capacity limits of tracks and vias. Existing EDA programs often have modules that can generate useful current density maps but can only make automatic changes to the layout if the voltage levels and component power demands are known. Thus, this part of the design process remains manual and relies heavily on the skill and experience of the designer to choose appropriate track widths and via aspect ratios. However, if this power consumption information could be made available to the artificial intelligence design assistant, this data could be synchronized with the layout software, so that machine-to-machine communication could be used to optimize the layout design automatically. Although such capabilities are not yet realized, ongoing advancements suggest they could become standard features soon.
Although a great deal of data can already be included in a cloud-based component database (for example, CELUS uses an enriched data block format that it calls CUBO™ to contain relevant information about a component application, such as signal mapping pin functionality, power supply requirements, etc. as well as any associated required components such as pull-up resistors, decoupling capacitors, crystals, etc. that are needed for full functionality), more data is often available in the individual component datasheet.
Hence the current focus on AI-assisted data mining to extract relevant data from both text and graphical information held in the datasheets. However, this process is not easy. Different manufacturers place equivalent information on different pages of their datasheets, so a data miner would need to work its way through all the text and graphs and recognize that, say, an efficiency figure given on page 1 of Manufacturer A’s datasheet is the same as the one given on the graph 2 on page 3 of Manufacturer B’s datasheet. Sometimes the information is simply missing and often the information is comparable but not directly equivalent.
For example, Manufacturer A might give an isolation withstand voltage of 3kVDC for one second, while Manufacturer B might specify 1kVAC for one minute. Which one is better? The answer often depends on the application and project definition. The task is extracting useful and valid data from datasheets thus requires expert knowledge artificial intelligence algorithms, capable of handling inconsistent data. However, as AI algorithms improve, so does the ability to extract and interpret data accurately, paving the way for comprehensive datasheet data mining functionality in the coming years. This evolving landscape underscores the transformative potential of AI in PCB design, promising continued innovation and efficiency gains for the industry as a whole.
Hence the current focus on AI-assisted data mining to extract relevant data from both text and graphical information held in the datasheets. However, this process is not easy. Different manufacturers place equivalent information on different pages of their datasheets, so a data miner would need to work its way through all the text and graphs and recognize that, say, an efficiency figure given on page 1 of Manufacturer A’s datasheet is the same as the one given on the graph 2 on page 3 of Manufacturer B’s datasheet. Sometimes the information is simply missing and often the information is comparable but not directly equivalent.
For example, Manufacturer A might give an isolation withstand voltage of 3kVDC for one second, while Manufacturer B might specify 1kVAC for one minute. Which one is better? The answer often depends on the application and project definition. The task is extracting useful and valid data from datasheets thus requires expert knowledge artificial intelligence algorithms, capable of handling inconsistent data. However, as AI algorithms improve, so does the ability to extract and interpret data accurately, paving the way for comprehensive datasheet data mining functionality in the coming years. This evolving landscape underscores the transformative potential of AI in PCB design, promising continued innovation and efficiency gains for the industry as a whole.
In conclusion, AI-assisted PCB design offers several significant advantages over traditional methods:
Speed and Efficiency: AI-powered design platforms streamline the design process by automating various tasks such as schematic generation, layout optimization, and component selection. This automation significantly reduces the time required to bring a product to market, enabling faster turnaround times and greater efficiency in design iterations.
Optimization and Performance: AI algorithms can analyze vast amounts of data to optimize designs for performance, reliability, and cost-effectiveness. By considering factors such as component specifications, signal integrity, and manufacturing constraints, AI-assisted designs can achieve higher levels of performance and reliability compared to manually crafted designs.
Enhanced Decision-Making: AI algorithms can assist engineers in making informed design decisions by providing real-time feedback and suggestions. This helps engineers identify potential issues early in the design process and explore alternative design options more efficiently, leading to better overall design outcomes.
Customization and Adaptability: AI-powered design platforms can adapt to the specific requirements of each project and user preferences. They can incorporate custom design rules, constraints, and preferences, allowing engineers to tailor designs to meet specific application needs while maintaining compatibility with industry standards and best practices.
Knowledge Transfer and Learning: AI-assisted design platforms can serve as valuable educational tools, especially for less-experienced engineers. By explaining design decisions, providing insights, and offering recommendations, AI systems can help engineers learn and improve their skills over time, contributing to professional development and knowledge transfer within organizations.
Risk Reduction: AI algorithms can help mitigate design risks by identifying potential issues, such as open or shorted connections and signal integrity problems before they become critical issues. This proactive approach to risk management can reduce costly design errors and rework, ultimately leading to more reliable and robust designs.
Overall, AI-assisted PCB design offers a transformative approach that combines the power of automation, optimization, and decision support to streamline the design process, enhance design outcomes, and drive innovation in the field of electronic design.
Applications
